"Dear friends, I am looking for detailed information about the Dutch East Indies plantation tokens of Tandjong Kassau. This tobacco plantation was located in the Residence East Coast of Sumatra, District Batoe Bahra, 10 miles away from the place Tebingtinggi Deli. The plantation has been in operation since 1890 by the “Sumatraanse Taback Gesellschaft Tandjong Kassau “ till c. 1908.

"The information known so far about these tokens is a incomplete description from an undated inventory list in the National Museum in Jakarta, as noted in my 2001 edition on these plantation tokens. This list includes tokens in denominations of 1 dollar, 50-25-10 en 5 cents in both nickel and tin, diameters 30.4-20.6 and 16.3 mm. respectively. Description: obverse: denomination in center, legend: SUM. TAB. GES. TANDJONG KASSAU Reverse: denomination.

"I would like to receive images and detailed information, weight, metal, stamp position, legend and/or denomination obverse and reverse, edge. Hopefully I can incorporate this new detailed information into my 7th edition on the Dutch East Indies plantation tokens.

"You can reach me by Email : lansenkil@hotmail.nl and/or : ajlansen@caiway.nl".

Can anyone help? -Editor

"The link you posted to the story of using AI to decipher charred ancient documents got me thinking. At one time I would have been amazed at the ability of computer systems to do this kind of thing, but now I'm a little more skeptical, based on a couple of recent experiences in particular.

"Last year a colleague in one of our numismatic clubs brought along a handwritten letter from the 1600s that he wanted to have transcribed. I volunteered to do it the old-fashioned way, having had a little experience in this area: carefully comparing letters in known words with shapes in words that were less certain, while another member offered to get an AI system to work on it. My text in the end was incomplete but marked uncertain areas as well as points where the text ran off the sides of the photocopied document; the AI version produced smooth English that read well but marked none of the missing areas ... and clearly didn't match the content of the original very well on closer inspection.

"More recently I was asked to decipher a text that was in Solitreo script, a variant of the Hebrew alphabet that I was unfamiliar with. Rather than try to match the handwriting myself using a chart, I asked ChatGPT to convert the text into standard Hebrew letters to make it more legible. It delivered a text with line-by-line numbering to help me compare the text ... but the number of letters per word, the number of words per line, and even the number of lines didn't match the source document.

"Both of these instances revealed a major pitfall of AI systems - they will guess based on context and favour something that reads well over something that's accurate."

Well, consumer-facing artificial intelligence today is built upon many separate tools working together behind the scenes; each individual tool has its limitations, and so does the part that orchestrates and presents the results of the tools. But academic researchers and other specialists use specific tools for specific problems. In the case of these charred papyrus scrolls, the project used AI as a high-powered vision tool to detect physical ink remnants that are invisible to the human eye, rather than attempting to generate missing remnants.

Since we're talking about AI, I thought I'd use it to flesh out my thesis. My prompt was "help me draft a response to this criticism of AI. Is it fair to compare deciphering charred ancient documents with these other use cases of AI?" -Editor

"That is a great point, and your examples perfectly illustrate why we have to be so careful with 'Generative AI.'

You’re spot on about the pitfall: systems like ChatGPT are essentially trained to be 'pleasers.' They prioritize fluency and helpful-sounding prose over raw data accuracy, which is a disaster for transcription where the gaps and uncertainties are the most important part. As you found with the 1600s letter, a 'smooth' read is actually a red flag if the source is damaged.

The difference with the charred scrolls (the Herculaneum project) is that they aren't using a chatbot to 'guess' the words. Instead, they use AI as a physical sensor to identify microscopic textures of ink on carbonized papyrus. It’s more like a super-powered X-ray than a writing assistant.

However, your skepticism is totally justified—whenever an AI 'interprets' a language it wasn't specifically built for (like Solitreo), it tends to default to 'making it look right' rather than 'being right.' Until we have models that are comfortable saying 'I don't know' or 'this is a gap,' the old-fashioned way is still the only way to ensure historical integrity."

That pretty much captures what I would have written myself in response. My phrasing would be different, but the core point remains. As always, reader thoughts and experiences are welcome, particularly in the context of numismatic research.

We're all still figuring out how to use and interpret the results of these new tools. They're very useful in certain applications, but must be applied with intention and have their results carefully reviewed. Yet sometimes they work remarkably well. I had a conversation recently with Roger Burdette, who'd been struggling for years searching for solutions to transcribe hand-written 19th century U.S. Mint records found in the National Archives, and today's solutions are getting very good. He even assisted a fellow researcher who'd located relevant documents in another language. With AI assistance, Roger transcribed the documents into text, then translated the results into English with very positive results, reducing what once would have been a weeks-long task to minutes. I'm still optimistic about the new-fashioned methods.

Roger provided some more background on his methodology, which provides careful guardrails around the underlying AI tool. Thank you. -Editor

"There is a substantive difference, however, between what I've developed for transcribing handwriting and Mr. Purdy's experiences. My software, called Natural Language Recognition (NLR), has been trained on both general word form and grammar models, and numismatic specific parameters combined with a long list of conditional characteristics. Together these allow very accurate transcriptions from "business handwriting" (generally post 1850), and from pre-1960 typewritten documents (before IBM Selectric typewriters) including correctly interpreting missing words, irregular spaces and some proper names. NLR also correctly interprets numismatic jargon such as "pyx" and "assay," etc. It does not retain original formatting - this was an intentional decision so that NLR could concentrate resources on language. NLR sits on top of GPT5 but uses the specialized training library and related parameters to work correctly. No such product is perfect and I review output for mistakes when appropriate.

"The Herculaneum scroll project makes high resolution phase contrast CT scans of the charred scrolls. Software developed by Brent Seales, a computer scientist at the University of Kentucky, uses certain "artificial intelligence" techniques to digitally reconstruct the scroll's spiral wound paper, unroll it, and then detect writing. This is extremely painstaking work and archive curators are very reluctant to allow access to the originals. There is an excellent PBS documentary 'Deciphering The Ancient Herculaneum Scrolls' available at https://www.youtube.com/watch?v=9z0SzSRAHTI."

Jeff Rock passed along word that he'll be speaking on "Numismatic and Philatelic Commemorations of the Queen of Crime" at the September 2026 Agatha Christie Festival in Torquay, England (her birthplace). Cool! -Editor

The talk takes a look at the coins, tokens, medals, paper money, and stamps issued to honour Agatha Christie, mostly in the half-century since her death. Like Christie's books, which were worldwide sensations, these collectible issues also span the globe. There will be a few surprises that have never been seen before - by neither coin collectors nor Christie specialists!

Having discovered the joys of Agatha Christie and coin collecting at the exact same time, at 8 years old, Jeff became a full-time coin dealer in the United States, specialising in the colonial-era coinage of North America, beginning when they were under British control. He also collects British tokens of the 18th century, further cementing his ties to England. Jeff says, 'It is an honour to give a talk that unites two of my most passionate interests.